Those cards, so familiar! Exactly what Opus produced for me.
Did Anthropic and/or OpenAI deliberately train their models to produce websites with a specific design language, or did these stylistic preferences emerge naturally as some kind of LLM-selected optimum?
It's not the base model, it's the system prompt in dev tools.
To give an example I'm personally frequently annoyed by, Google's Antigravity will consistently use the word "anthropomorphic" while "thinking" and the end result will consistently have obnoxiously large border radius (kind of like Android's design language).
Codex on the other hand likes to make websites with blue elements on a black background and likes to use emojis for icons for some reason, which is a terrible idea accessibility-wise.
A "little wobble" may not change it more than a "little", but enough "little wobbles" over time become a "big wobble" that may change it in a "big" way. The right question to ask is: what's the elasticity? So far the elasticity of domestic migration to tax increases has been smaller than many expected due to network effects and inertia, but nevertheless if you look at the population growth rates of high tax states like CA and NY and compare it to low tax states like FL and TX, you will definitely see a pattern. Rational people think on the margin. Perhaps only a few people will move if you increase tax rates by 0.1%, but more on average will decide to move than if you hadn't raised taxes - the question really is, how many?
Interesting to note, not sure if this was known publicly before today's blog post:
Rewriting Bun with dynamic workflows
An example of what dynamic workflows can unlock at scale is the recent rewrite of Bun. Jarred Sumner used dynamic workflows to port Bun from Zig to Rust with 99.8% of the existing test suite passing, roughly 750,000 lines of Rust, and eleven days from first commit to merge. One workflow mapped the right Rust lifetime for every struct field in the Zig codebase. The next wrote every .rs file as a behavior-identical port of its .zig counterpart, hundreds of agents working in parallel with two reviewers on each file. A fix loop then drove the build and test suite until both ran clean. After the port landed, an overnight workflow addressed unnecessary data copies and opened a PR for each for final review. While not yet in production, all of this was handled by dynamic workflows. Jarred will be writing about this more in the future.
I'm extremely skeptical that dynamic workflows had anything to do with this. I've been able to refactor one of the most complicated parts of our code base with similar results.
Mechanical refactors are relatively straight forward for agents.
> I've been able to refactor one of the most complicated parts of our code base with similar results. Mechanical refactors are relatively straight forward for agents.
A rewrite of bun in Rust is unlikely to be a trivial mechanical refactor. And if you are not sharing what the complicated parts were, or how big it is, how do we assess that the task was similar?
Unless you are intimately familiar with the bun codebase and you've already made that assessment.
> I dunno, 1.3% seems.. not at all unreasonable? I live in TX and that's about what my property taxes are, for a property valued at several orders of magnitude less than any of Ken Griffin's NYC properties.
Bear in mind, it's 1.3% on top of the existing ~1.8% average NYC property tax rate, so it may still be comparatively expensive relative to TX property taxes.
> In the UK, the equivalent tax on housing is council tax, and local councils in Great Britain (but not Northern Ireland) are empowered to double the rates of council tax on second homes.
Very interesting to know. Many readers may not be aware that council tax in the UK is quite regressive and tops out at ~£4-5K / year on properties valued higher than ~£1M. So you can own a £5M GBP house and still pay only £5K / year for an annual effective property tax rate of just 0.1%.
This is one of the reasons buying a luxury house in the UK is comparatively quite cheap in terms of total cost of ownership compared to many states in the US where you have to pay much higher property tax rates, insurance, and so on.
So even if the council tax is doubled on a second home, you still might be paying only 0.2%. Compare that to an average property tax rate of ~1.8% in NYC (before pied-a-terre).
Yeah thanks for nothing for comparing a single kind of tax to your country, whilst your country/states don't have the excessive overall tax regimes as are present in Europe.
Nothing, absolutely nothing do we have to adjust to America, neither up or downwards.
That being said, and as much as I think Mamdani is an Ideologue, taxing second, unoccupied homes sounds absolutely reasonable (at least if they aren't rented out). Expect all kinds of shenanigans to circumvent this, but still.
> Why? I don't need 10x more stuff. I'd far rather spend 10x less time working.
It's a free market - if you can find someone willing to hire you to work 0.5 days a week, and are happy receiving the same income you do today, you can do so.
But your living standards won't improve, while everyone else's will, and by 10x.
And the people you are competing with to buy housing, a car, consumer goods, healthcare, food, and everything else - people who want to work 5 days a week - will have 10x more money than you. Resources like housing, which are supply constrained, will seem to go up dramatically in price relative to your income, and you'll be living in a slum.
> Resources like housing, which are supply constrained, will seem to go up dramatically in price relative to your income, and you'll be living in a slum.
That sounds like a problem with that system. If we're 10x more productive why can't we make 10x as much nice housing?
> That sounds like a problem with that system. If we're 10x more productive why can't we make 10x as much nice housing?
Because the supply of land is relatively inelastic (and to some extent building materials and construction labor). There isn't suddenly 10x more land on the planet because we all start using AI. Maybe we can make more land or build taller buildings, but it gets progressively more difficult and expensive. And because of NIMBY of course.
Effective working hours are not set by absolute productivity - they are set by an equilibrium between two forces:
1. Competitive market dynamics. If you only work four days a week, other employees and companies who are willing to work five days a week will do so and get ahead of you, and you are more likely to get fired or to go out of business. This force pushes us all to work longer (and harder) so we have more money to enjoy in our leisure time.
2. A society's willingness to sacrifice days of leisure for days of work. There are only seven days in a week. The tradeoff between work and leisure - production and consumption - is ultimately what determines how hard we all work. This force pushes us all to work less so we have more time to spend our money.
Economists think on the margin. It's easy to demonstrate these two principles to yourself by thinking through worked examples from different starting points.
Whether the equilibrium lands at 2 days of work to 5 days of leisure, or 5 days of work to 2 days of leisure, depends on our collective preferences, which vary between countries and cultures but have tended to be relatively durable over time.
No technology so far has shifted this balance much - not the steam engine, the industrial revolution, the invention of the personal computer, the internet - and there's no reason to believe "AI" will be any different.
The logical conclusion of this is that - assuming we're all 10x more productive - we'll still be working 5 days and enjoying 2 days a week, but we'll consume 10x more, or everything we consume will be 10x higher quality. Hardly a bad thing.
> assuming we're all 10x more productive - we'll still be working 5 days and enjoying 2 days a week, but we'll consume 10x more, or everything we consume will be 10x higher quality
Who is "we all"? To me, it sounds like the relative few who happen to have those jobs that have the 10x productivity boost but also receive the monetary upside (via ownership).
The rest of the hard-to-automate jobs will likely see their wages crater as the workers whose jobs got automated flood those labor markets - i.e. office worker turned skilled physical laborer.
This will further enrich the previous small group relative to the masses, as they will pay lower prices and receive higher quality goods and services due to competition between everyone else. Prices will fall not by miraculous AI robots but by squeezing labor.
This is the scenario - neofeudalism - that may await us absent strong mechanisms to replace the broad productivity redistribution the social technology known as "jobs" provided. Hardly a good thing.
You’re speculating about the effects of AI specifically but responding to a comment is about productivity generally. Historically we’ve seen massive productivity gains, and yet we’re all still here working.
First sensible take I've seen in this thread. People especially seem to forget the consequences of #1: The company where everyone is working 0.5 days a week will almost certainly get outcompeted, very quickly, by the company where everyone is working 5 days a week. In fact company #2 can probably precompute company #1 even if they have much lower quality staff on average.
Attraction towards equilibrium is real, but that equilibrium is not a given. It comes from intrinsic biological constraints, individual preferences, cultural norms, ideology, habits and expectations. Even fixing individual preference, the aggregate preferences is structurally mediated (collective or individual bargaining, labour laws). Through all those factors there's path dependency and friction towards equilibrium.
The point is: "a society's willingness" is doing a _huge_ amount of work in that framing. This willingness is precisely what is up for debate when we discuss work days.
The whole of humanity is one big system of self-feedbacks. Equilibria are only reached with respect to constraints (otherwise there is only one equilibrium, which is heat death!). The more you zoom out, the more "givens" come up for their own analysis.
Did Anthropic and/or OpenAI deliberately train their models to produce websites with a specific design language, or did these stylistic preferences emerge naturally as some kind of LLM-selected optimum?
reply